Just kidding. I never found it boring.
Documentation is critical for any organization intent on building its institutional knowledge. It’s the ledger of your org’s impact, history and DNA: whether that’s an internal written policy, technical documentation, annual report, style guide or something else entirely.
But documentation is also, often, the very last thought in the technology process. It’s that task you know you should do, may even try to do, but usually don’t get around to because it simply isn’t priority.
So we all just tell ourselves that as long as the info lives in somebody’s head, our orgs will be fine.
Enter AI, which presents a new and unique opportunity for all of us to do better. First, AI can be great for getting new documentation done at all. Generative AI can help produce thorough & useful reference material for your org, with the right prompting and human review. Those two caveats are deserving of their own blog post, though I’ll resist the urge today.
But going further, AI can also help you decipher and act on existing documentation. Put another way, docs provide context that can take your AI use to the next level.
For those who don’t understand this as a technical term, context is the information/memory/data that AI uses to generate relevant responses. Anything you type into an AI chat directly, or the files you link in a chat/project/knowledge base/wherever, all act as context that tells the AI what you think it needs to know,
And for automating new processes with AI, docs also play a foundational role in helping create the instructions for those AI agents. (AI prompting and instruction-building are technically separate work streams, but the potential is similar.)
So stop treating your org docs as institutional weight, or that task that doesn’t serve an emergent need. Instead think of your org’s docs as a strategic well of context for your chosen AI. And if your org’s doc maintenance is top notch…well, that’s where the utility potential can really ramp up.
The stronger your org’s documentation muscle, the more tailored & useful AI becomes.
Before you go dumping all of your docs into ChatGPT..
Or Claude, Perplexity, or whatever your team’s AI tool of choice is these days!
Like with all prompts, consider your data-sharing agreements when deciding what information to give to AI.
For example… if your team’s data may be used to train models, then you likely don’t want to share intellectual property or other proprietary info about your org. (In this scenario, you also don’t want to enter anyone’s personal or sensitive data either – out of respect for their data privacy. )
You also want to consider the relevance of the information you’re sharing. AI won’t know that an aspect of your process is out of date, or that one of the programs outlined in your onboarding manual is slated to retire, or that you no longer use certain language when referring to constituents. It’s going to deem whatever information you give it as relevant, even the bad info, because you’re providing it.
Think about it like onboarding a new coworker, but giving them an org manual from 2017. At best, that’s not useful. At worst, it can be harmful to your impact.
How AI makes existing documentation more useful
The list below is not exhaustive. There are many, distinct ways that you – or another person at your org – can extract unique value by using docs alongside AI.
But the examples below are meant to give a feel, and generally share the same efficiency benefit. Using documentation as context 1) saves time with our AI prompts AND 2) contributes to more relevant AI outputs.
Think about your experience in an AI chat. Each time, you provide contextual info to focus the output: details like what you want to see, the format, and the background info that needs to be considered in relation to performing that task to your satisfaction.
Documentation can help spare you some of that initial explanation AND post-output corrective prompting. Put another way, this can help you work more efficiently and consistently.
That said, on to the examples.
p.s. I think we all know this, but just in case. The ideas below are in service of supporting your human staff. They are not replacements for the work and decisions driven by your peeps. (Remember that AI can never know your org like your people do, and no amount of stellar documentation changes that. But still aim for stellar docs.)
6 Distinct Ways to Leverage Docs with AI
1) Reproduce tone, style & branding
If your org has a blog, impact report or even a style guide that breaks down “how” your org presents itself to the world – this can provide the base for AI instructions to help generate new content, whether that be for internal use or external consumption. (A real person should still hold final responsibility for reflecting your org’s brand & voice in materials.)
2) Optimize processes & workflows
This has been one of my favorite, new ways to experiment with AI. Workflow diagrams are incredibly useful for visualizing the steps in a process and seeing them in relation to the whole. You can always use AI to build these if it’s too daunting to do yourself, but you can also use AI to examine those diagrams and explore ways to optimize. In your prompt, provide extra narrative on which steps in that flow tend to cause bottlenecks or issues – or where you want to see different results – and see what ideas AI has to offer you.
3) Optimize data/tech architecture
Similar to the last point – data model diagrams, data dictionaries, dev philosophies or any other technical documentation can be good fodder for an AI-assist on your tech approach. (But let your trusted tech person be the final judge on how actionable those recs really are.)
4) Unearth follow-up tasks
If there’s one task harder to prioritize than creating docs, it’s updating them. So if you’re not sure which policies could use a refresh, or whether it’s time to update aspects of your org strategy or research, an AI scan could provide useful recs and structure – including what to tackle and a plan to help you execute those updates.
5) Translate reports or SOPs to new AI-driven processes
We briefly covered this in the previous section. For processes that make sense to automate with generative AI (not all of them will), docs can provide a helpful base for creating specific, AI-friendly instructions. You can then plug those final instructions into your Claude Skills, Gemini Gems or any of the other LLM equivalents.
Tip: The point of these instructions is so you can call them into future prompts, without having to spell out the specifics each time. For repeatable low-risk tasks, this can be a helpful timesaver.
6) Use AI to document AI
Documentation is an important driver of accountability. So if we’re adding flashy new tech to the stack (one that creates more management work at the moment than it saves..) then AI might as well help lighten the load!
So consider how you might use AI as a self-auditing tool, particularly for your higher-risk/reward processes. Ask for recaps of chats. Brainstorm a structure for those recaps (i.e. instructions) that helps your org see its AI use more clearly. Consider creating an “AI log” system, where a human reviews those recaps and a final log file gets saved to your document repository. This can help with understanding your org’s AI ROI (return on investment). But more importantly, it can provide crucial visibility into where your human-in-the-loop approach sits / needs to be beefed up.
Wrapping Up + Tips to Start Today
AI is giving the need for documentation new life. If you ever needed a reason to be diligent about documenting your processes, technology and org artifacts, that reason has arrived.
So if you’re sold on getting started, start small. First, audit those considerations I mentioned above – like your doc relevance and data-sharing agreements – before anything else.
Next, pick a document or specific task you want to try. When you’re just starting out, the best AI experiments are your low-hanging fruit: tasks that don’t carry much risk if they go wrong, don’t require a ton of upfront lift, but do carry a tangible benefit if done well.
And if you’re strapped for time, set a timer for how long you’re willing to give AI a chance to show what it can do. It’s easy to get sucked in to prompting and re-prompting as you engage in “conversation” with your AI chat (part of the design of these tools, unfortunately). Constraining your time in a single sitting helps keep you honest — and doesn’t mean you can’t set another session trial for another time.
Leave a Reply